Skip to content

Run cleanup and honor return_dict for SDXL latent output #14680

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
Devam0311 wants to merge 1 commit into
base: main
Choose a base branch
from
Open

Run cleanup and honor return_dict for SDXL latent output #14680

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
17 changes: 10 additions & 7 deletions src/diffusers/pipelines/controlnet/pipeline_controlnet_inpaint_sd_xl.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1866,16 +1866,19 @@ def denoising_value_valid(dnv):
if not output_type == "latent":
image = self.vae.decode(latents / self.vae.config.scaling_factor, return_dict=False)[0]
else:
return StableDiffusionXLPipelineOutput(images=latents)
image = latents

# apply watermark if available
if self.watermark is not None:
image = self.watermark.apply_watermark(image)
if not output_type == "latent":
# apply watermark if available
if self.watermark is not None:
image = self.watermark.apply_watermark(image)

image = self.image_processor.postprocess(image, output_type=output_type)
image = self.image_processor.postprocess(image, output_type=output_type)

if padding_mask_crop is not None:
image = [self.image_processor.apply_overlay(mask_image, original_image, i, crops_coords) for i in image]
if padding_mask_crop is not None:
image = [
self.image_processor.apply_overlay(mask_image, original_image, i, crops_coords) for i in image
]

# Offload all models
self.maybe_free_model_hooks()
Expand Down
17 changes: 10 additions & 7 deletions src/diffusers/pipelines/controlnet/pipeline_controlnet_union_inpaint_sd_xl.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1878,16 +1878,19 @@ def denoising_value_valid(dnv):
if not output_type == "latent":
image = self.vae.decode(latents / self.vae.config.scaling_factor, return_dict=False)[0]
else:
return StableDiffusionXLPipelineOutput(images=latents)
image = latents

# apply watermark if available
if self.watermark is not None:
image = self.watermark.apply_watermark(image)
if not output_type == "latent":
# apply watermark if available
if self.watermark is not None:
image = self.watermark.apply_watermark(image)

image = self.image_processor.postprocess(image, output_type=output_type)
image = self.image_processor.postprocess(image, output_type=output_type)

if padding_mask_crop is not None:
image = [self.image_processor.apply_overlay(mask_image, original_image, i, crops_coords) for i in image]
if padding_mask_crop is not None:
image = [
self.image_processor.apply_overlay(mask_image, original_image, i, crops_coords) for i in image
]

# Offload all models
self.maybe_free_model_hooks()
Expand Down
17 changes: 10 additions & 7 deletions src/diffusers/pipelines/pag/pipeline_pag_sd_xl_inpaint.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1752,16 +1752,19 @@ def denoising_value_valid(dnv):
if needs_upcasting:
self.vae.to(dtype=torch.float16)
else:
return StableDiffusionXLPipelineOutput(images=latents)
image = latents

# apply watermark if available
if self.watermark is not None:
image = self.watermark.apply_watermark(image)
if not output_type == "latent":
# apply watermark if available
if self.watermark is not None:
image = self.watermark.apply_watermark(image)

image = self.image_processor.postprocess(image, output_type=output_type)
image = self.image_processor.postprocess(image, output_type=output_type)

if padding_mask_crop is not None:
image = [self.image_processor.apply_overlay(mask_image, original_image, i, crops_coords) for i in image]
if padding_mask_crop is not None:
image = [
self.image_processor.apply_overlay(mask_image, original_image, i, crops_coords) for i in image
]

# Offload all models
self.maybe_free_model_hooks()
Expand Down
17 changes: 10 additions & 7 deletions src/diffusers/pipelines/stable_diffusion_xl/pipeline_stable_diffusion_xl_inpaint.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1714,16 +1714,19 @@ def denoising_value_valid(dnv):
if needs_upcasting:
self.vae.to(dtype=torch.float16)
else:
return StableDiffusionXLPipelineOutput(images=latents)
image = latents

# apply watermark if available
if self.watermark is not None:
image = self.watermark.apply_watermark(image)
if not output_type == "latent":
# apply watermark if available
if self.watermark is not None:
image = self.watermark.apply_watermark(image)

image = self.image_processor.postprocess(image, output_type=output_type)
image = self.image_processor.postprocess(image, output_type=output_type)

if padding_mask_crop is not None:
image = [self.image_processor.apply_overlay(mask_image, original_image, i, crops_coords) for i in image]
if padding_mask_crop is not None:
image = [
self.image_processor.apply_overlay(mask_image, original_image, i, crops_coords) for i in image
]

# Offload all models
self.maybe_free_model_hooks()
Expand Down
11 changes: 6 additions & 5 deletions src/diffusers/pipelines/stable_diffusion_xl/pipeline_stable_diffusion_xl_instruct_pix2pix.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -960,13 +960,14 @@ def __call__(
if needs_upcasting:
self.vae.to(dtype=torch.float16)
else:
return StableDiffusionXLPipelineOutput(images=latents)
image = latents

# apply watermark if available
if self.watermark is not None:
image = self.watermark.apply_watermark(image)
if not output_type == "latent":
# apply watermark if available
if self.watermark is not None:
image = self.watermark.apply_watermark(image)

image = self.image_processor.postprocess(image, output_type=output_type)
image = self.image_processor.postprocess(image, output_type=output_type)

# Offload all models
self.maybe_free_model_hooks()
Expand Down
19 changes: 19 additions & 0 deletions tests/pipelines/stable_diffusion_xl/test_stable_diffusion_xl_inpaint.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -692,6 +692,25 @@ def test_stable_diffusion_xl_inpaint_2_images(self):
image_slice2 = images[1, -1, -3:, -3:]
assert (image_slice1 - image_slice2).abs().max() > 1e-2

def test_latent_output_runs_cleanup_and_honors_return_dict(self):
# Regression: `output_type="latent"` returned `StableDiffusionXLPipelineOutput(images=latents)`
# immediately, which skipped `maybe_free_model_hooks()` and ignored `return_dict=False`.
sd_pipe = self.get_pipeline()

called = {"cleanup": False}
sd_pipe.maybe_free_model_hooks = lambda: called.__setitem__("cleanup", True)

inputs = self.get_dummy_inputs()
inputs["output_type"] = "latent"
inputs["return_dict"] = False
output = sd_pipe(**inputs)

assert called["cleanup"], "`maybe_free_model_hooks()` should still run for latent output."
assert isinstance(output, tuple), "`return_dict=False` should return a tuple."
assert torch.is_tensor(output[0])
# Raw latents keep the VAE latent channel count; postprocess would have produced 3 channels.
assert output[0].shape[1] == sd_pipe.unet.config.in_channels

def test_pipeline_interrupt(self):
sd_pipe = self.get_pipeline().to(torch_device)

Expand Down
Loading